The privacy of on-premise. None of the procurement.
On-premise AI is the right answer for a 200-lawyer firm with an IT budget.
Muet is the right answer for everyone smaller.
What on-premise actually means
The catch is everything around the model. Hardware procurement, network isolation, model deployment, RAG configuration, identity and access management, monitoring, backups, model upgrades when better open-source weights ship every quarter, and someone in IT who keeps it running.
What on-premise costs to actually deploy
- Hardware. A workstation or small server capable of running a 70B-class model: roughly $5,000–$30,000 depending on whether you buy a Mac Studio cluster or a GPU server.
- Setup. 8–12 weeks from procurement to production for a mid-size firm: 2–3 weeks for hardware and network, 2–3 weeks for model and RAG, 2–4 weeks of testing, 2 weeks for training and rollout.
- Staffing. One Linux- and container-fluent administrator can manage the infrastructure for up to 200 users; smaller firms typically don’t have that person on staff.
- Ongoing cost. Power, cooling, model updates, security patches, and the fact that the state of the art in open-source models is moving every few months and a year-old deployment is meaningfully behind.
What Muet costs to deploy
- Hardware: the Mac you already own.
- Setup: the time it takes to drag the .dmg to Applications.
- Staffing: none.
- Ongoing cost: $390 per device per year. Updates are automatic.
- Time to first useful output: about ten minutes from download to first answered question.
The privacy comparison
- On-premise. Documents live on the firm’s servers. Multiple users can access them. The threat surface is the firm’s network, the IT staff with privileged access, and the audit trail of who looked at what.
- Muet. Documents live on one user’s Mac. Nobody else, not us, not the firm’s IT, has access. The threat surface is the laptop, the device policy, and the user.
For a firm doing matter-team review with twelve associates touching the same data room, on-premise is closer to how the work is already organised. For one lawyer drafting a privileged memo on their own machine, Muet is closer.
The two often run alongside each other
- On-premise for matter teams collaborating across thousands of documents in a shared workspace, with the firm’s identity, audit, and review controls wrapped around it.
- Muet on partner and associate laptops for the work that doesn’t belong in the shared workspace at all: pre-engagement pitches, internal investigations, partner-level HR matters, off-the-record work, personal correspondence with clients.
The firm gets the on-premise deployment’s collaboration story and Muet’s laptop-level privacy story without either compromising the other.
Where each makes most sense as the only AI tool
- On-premise alone.Firms processing >2M tokens/day of confidential text, government and defense work with hard air-gap requirements, or regulated facilities where every endpoint is centrally managed. Muet doesn’t try to replicate hardened-network controls.
- Muet alone. Solos, 2–25 lawyer firms, in-house counsel teams, and any practice without a dedicated IT budget. The same privacy posture, none of the deployment cost.
- Distributed teams where every lawyer wants the tool on their own machine, including the ones travelling or working from home. One license per device, no central deployment.
Try it
Pricing
30-day money-back guarantee. Apple Silicon (M1 or newer), macOS 14 or newer. One licence per device, easy to expense.